Wednesday  |  Conference Center B  |  03:10 PM–03:30 PM

#13706, SerialTrack: ScalE and Rotation Invariant Augmented Lagrangian Particle Tracking

Particle tracking is of key importance for quantitative analysis of dynamic processes from time-lapse image data, where individual particles or fiducial markers can be automatically detected and tracked. There are various existing approaches to track individual particles, i.e., single particle tracking (SPT), ranging from simple nearest neighbor searches to solving global optimization problems, including topology-based particle tracking approaches. Among these methods, the nearest neighbor searching algorithm is suitable for relatively low numbers of particles that undergo displacements smaller than the typical interparticle separation distance, while the topology-based particle tracking algorithms reconstruct large, complex motion fields but favors large particle numbers. These two methods are computationally efficient; however, both have limitations on the particle seeding densities. In addition, every single particle is tracked by only using its local information, which no guaranty that the final tracked motion fields are kinematically admissible for a given medium. Global optimization particle tracking methods can guarantee the uniqueness and kinematic admissibility of the tracked motion, but they are more computationally expensive.

Here we present a new tracking algorithm to take advantage of both local and global tracking methods, which we call the ScalE and Rotation Invariant Augmented Lagrangian Particle Tracking (SerialTrack) method. This method builds a scale and rotation invariant topology-based feature for each particle, iteratively within a multiscale tracking algorithm. The global kinematic compatibility condition is applied as a global augmented Lagrangian constraint to enhance the tracking accuracy. Using both synthetically and experimentally generated 2D and 3D images including homogeneous and heterogeneous non-affine deformation fields in soft materials, complex fluid flows, and cell-generated deformations, we show that our new SerialTrack method can track both sparse and dense particles and resolve large and complex motion fields with high accuracy and computational efficiency. SerialTrack and other particle tracking methods are also quantitively compared with other full-field measurement techniques including 2D Digital Image Correlation and 3D Digital Volume Correlation to compare their performance. We maintain an open-source MATLAB implementation that is freely available to download (https://github.com/FranckLab).

Jin Yang   University of Wisconsin-Madison
Matthew Fu   California Institute of Technology
Yue Yin   Carnegie Mellon University
Alexander Landauer   National Institute of Standards and Technology
Selda Buyukozturk   Brown University
Jing Zhang   University of Wisconsin-Madison
Luke Summey   University of Wisconsin-Madison
Alexander McGhee   University of Wisconsin-Madison
John Dabiri   California Institute of Technology
Christian Franck   University of Wisconsin